From: Mike H. <mho...@gr...> - 2003-09-17 20:12:39
|
I don't actually think it's possible to write a sort-middle SPU. The problem is that in a sort-middle architecture like the IR2, the primitives are submitted, then the system load balances fragment generation, and sorts those into the rasterizes. Using hardware systems, there is no way to get access to the fragments before they have exited the entire pipeline, ie gone through the raster stage. You might be able to do this purely in software by hacking Mesa into several pieces, but why would you want to use software rendering? You can sort of get to this by doing a sort-first-last hybrid in which you sort the geometry out to various servers (tilesort) and then do a readback and composite on the servers (readback/binaryswap/SLIC/etc). However, you are really offloading the geometric load, fragment load, and raster load onto multiple systems. This is really a load balancing setup and not sort-middle. -Mike David wrote: > Hi, > > > > I am currently doing some research on heterogeneous parallel > architectures (in particularly different sorting techniques), and was > about to write a sort-middle SPU in an attempt to replicate the > Pomegranate architecture. > > > > However, I am currently running Chromium 1.4 on an SGI machine > containing infinite reality2 graphics pipes, and this supports > sort-middle natively. So was wondering, is the system already > performing sort-first, sort-middle and sort-last rendering techniques. > As this would mean that I would have to explicitly write a sort-middle > SPU. > > > > If anyone has any suggestions I would be most grateful. > > > > David > |